-8.1 C
New York

AI is now capable of using a computer by itself, but creators have taken measures to prevent a potential disaster

Published:

The impact of artificial intelligence has undoubtedly managed to change the current paradigm of technology. Just in the last few days, we have had all kinds of news related to this technology, as it has become the central focus of Apple’s future iPhones and, at the same time, was the protagonist of a very bizarre story that took place in a high school. However, few updates have been as ambitious as the latest one announced by Anthropic, one of the most relevant companies in the sector.

In my new favorite RPG, Gandalf has sent me to therapy and even made Obi Wan turn to the dark side. I'm addicted to a chat with AI

As noted by the The Verge portal in a recent publication, Anthropic launched the “computer usage” feature on their Claude 3.5 Sonnet model. In this way, AI is now capable of controlling a computer similarly to humans, as it can see the screen, move the cursor, click, and even type. The feature, currently available in public beta through the API, already allows developers to test it in their applications. And, according to the first curious users, this tool is ahead of similar features from Microsoft, Google, and OpenAI.

The era of autonomy arrives for artificial intelligence

Anthropic itself has indicated that this feature is still experimental and, therefore, may be prone to errors. Thus, Claude 3.5 uses a technique called “flipbook” that allows not only to observe the screen but also to take screenshots to learn, but this situation may lead to the loss of quick actions or brief notifications. Additionally, the language model has several restrictions: it cannot engage in social networks, it must not involve itself in electoral issues, and it is forbidden from registering web domains.

This way, the new version of Claude 3.5 offers significant improvements in encoding benchmarks and tool usage. In these tasks, it has been able to improve its performance in SWE-bench from 33.4% to 49%, surpassing other public models in the market. And, despite the improvements, the company maintains the same price and speed for Claude 3.5 Sonnet, a situation that makes this model much more competitive than its predecessors. With it, the era of autonomy has already arrived for artificial intelligence.

In 3DJuegos | An Apple study delves into what the main issue of artificial intelligence is and how it affects its performance

In 3DJuegos | The new trend among hackers is invisible texts for humans that manage to deceive AI to steal private data

Main image by Louis Reed (Unsplash)

Related articles

spot_img

Recent articles

spot_img